One year after the European Accessibility Act (EAA) became enforceable, the Netherlands has a case study in the gap between compliance-on-paper and compliance-in-practice. Research agency Level Level, which audits the country's 300 largest webshops annually, found that automated accessibility errors on homepages fell from an average of 108 per site in 2025 to 25.7 in 2026 — a drop of more than 75% (Bright.nl). That is a genuinely large improvement, and it happened in roughly the twelve months since the EAA's 28 June 2025 application date (Directive (EU) 2019/882).
But when Level Level moved past automated scanning and had real users test the 15 largest retailers end-to-end, the picture reversed. Only two of those fifteen sites let a screen-reader user complete a purchase independently and without barriers; thirteen presented significant obstacles during checkout, and at four, ordering without help was simply impossible (Emerce; NOS). Retailers cleaned up the markup an automated scanner can see — alt text, contrast, ARIA labels — while the actual purchase flow, the part a scanner struggles to test end-to-end, stayed broken for most disabled shoppers.
The regulator's case, stated fairly
The Dutch Authority for Consumers and Markets (ACM), which has enforced EAA compliance for e-commerce since June 2025, has its own data point reinforcing the same story. Its own audit of roughly 100 of the country's largest webshops and telecom/energy sites, published 24 March 2026, found 61% not digitally accessible and another 33% seriously impeded — meaning, combined, ordering was unreliable or impossible at roughly 19 in 20 of the sites tested (ACM). ACM visited the ten worst performers directly to explain what needed fixing, and has said it will re-test in three months, with fines on the table for whoever still falls short (NOS).
The steelman for this approach is straightforward and worth taking seriously. Roughly 4.5 million Dutch residents live with a disability that can affect how they use the web, and for over a decade digital accessibility has been treated as a best practice to adopt when convenient rather than a baseline to meet by default. A law with teeth changed retailer behavior faster than a decade of voluntary WCAG guidance did — the 75% error reduction is proof the incentive worked. And a regulator that names specific companies, gives a fixed re-test window, and threatens an actual fine is doing exactly what proportionate enforcement should look like: warn first, target the worst offenders, escalate only if they ignore the warning.
Where the compliance model breaks down
The honest complication is that the EAA — like most accessibility law — is written and largely audited against machine-checkable technical standards (EN 301 549, itself built on WCAG success criteria). Those criteria are necessary but not sufficient: a page can pass every automated check and still fail a screen-reader user at the one step that matters, completing a purchase. Level Level's own comparison makes that gap unusually visible — a 75-point improvement on the metric regulators and vendors can score automatically, against almost no movement on the metric that determines whether a disabled customer can actually buy something.
That mismatch is a genuine policy design problem, not just a retailer failure. If ACM's three-month re-test relies primarily on the same kind of scan that produced the 108-to-25.7 improvement, retailers will optimize for the test again — polishing markup rather than fixing checkout flows — and the compliance numbers will look better while the lived outcome barely moves. Regulators serious about outcomes need to weight practical usability testing, not just automated conformance scores, in whatever counts as compliance for the fine threshold. ACM's decision to physically visit the ten worst performers is a step in that direction; it should extend the same scrutiny — real users attempting real purchases — to the re-test in three months, not just a rerun of the scanner.
The proportionate read
None of this argues against the EAA or against ACM's enforcement plan. A one-year, 75% reduction in detectable errors, achieved without the retail sector collapsing or major webshops exiting the Dutch market, is exactly the kind of outcome proportionate regulation is supposed to produce: real behavior change, calibrated timelines, warnings before fines. The risk is not that the law is too strict — it's that its metrics are too easy to satisfy without fixing the thing it was written to fix. ACM should hold that line at the three-month mark by testing purchases, not just pages, before it decides who gets fined.